SIGMAP: an explainable artificial intelligence tool for SIGMA-1 receptor affinity prediction.

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and medicinal chemistry
• Artificial intelligence in pharmacology

Background:

• Sigma-1 receptor (S1R) modulators show therapeutic potential for neurodegeneration, cancer, and viral infections like COVID-19.
• Accurate *in silico* prediction of S1R affinity is essential for efficient drug design.
• Existing methods require improvement for reliable S1R modulator identification.

Purpose of the Study:

• To develop and validate accurate computational models for predicting S1R modulator activity.
• To create a user-friendly platform for medicinal chemists to aid in rational drug design.
• To leverage explainable AI (XAI) for better understanding of predictive model decisions.

Main Methods:

• A curated dataset of 25,000+ S1R small molecule bioactivity data points was extracted from ChEMBL v33.
• Twenty-five classifiers were trained using five distinct molecular fingerprints and various machine learning algorithms.
• Explainable AI techniques, including SHAP and Contrastive Explanation, were applied to interpret model predictions.

Main Results:

• Most developed classifiers exhibited good predictive performance.
• The top-performing model, utilizing a support vector machine with Morgan fingerprints, achieved an AUC of 0.90.
• A user-friendly web platform, SIGMAP, was developed to provide access to the best predictive model.

Conclusions:

• The developed computational models, particularly the best-performing SVM model, demonstrate high accuracy in predicting S1R affinity.
• The SIGMAP platform, incorporating XAI, offers a valuable tool for medicinal chemists in the rational design of novel S1R modulators.
• This approach facilitates the discovery of new therapeutics for S1R-related conditions.